The conference on Neural Information Processing Systems, under its new acronym NeurIPS, is one of the premier annual AI conferences in the world and a top venue for publication of machine learning research. The conference has grown significantly since it began in 1987; in 2018 almost 9,000 people will travel to Montréal, Canada for the conference and 4,854 papers were submitted for review.
The process to choose papers and workshops is highly selective. We are delighted that the Turing has a large number of researchers attending and many papers accepted in such a competitive environment.
Turing papers which were accepted this year cover topics such as Geometrically Coupled Monte Carlo Sampling (Adrian Weller - Programme Director for Artificial Intelligence, Turing Fellow, University of Cambridge); a Bayes-Sard Cubature Method (Chris Oates - Data-Centric Engineering Group Leader, Turing Fellow, Newcastle University); and "neurosymbolic" program synthesis for lifelong learning (Charles Sutton – Turing Fellow, University of Edinburgh), to name only a few.
Workshops are also a significant part of the NeurIPS conference, providing a forum for researchers to discuss important, topical research questions and challenges. Competition is high and of 114 workshop proposals, only 39 were selected by the judging panel. Excitingly, of those 39, three of the workshops involve Turing Research Fellows as organisers. They are:
Machine Learning for Molecules and Materials – sponsored by the Turing
Turing co-organisers: José Miguel Hernández-Lobato, Matt Kusner, Brooks Paige
Privacy Preserving Machine Learning – co-sponsored by the Turing
Turing co-organisers: Adria Gascon, Adrian Weller
Critiquing and Correcting Trends in Machine Learning
Turing co-organisers: Matt Kusner, Brooks Paige, Yee Whye Teh
We heard from two of the Turing workshop co-organisers before they headed off to Montreal last week.
Adria Gascon (University of Warwick) Turing Research Fellow and co-organiser of Privacy Preserving Machine Learning, said:
“I’m co-organising a workshop on privacy-preserving machine learning at NeurIPS. The general goal of the workshop is to offer a venue to unify different research perspectives to the problem of making inferences on private data. These include cryptographic approaches based on techniques such as Multi-Party Computation and Homomorphic Encryption, statistical approaches like Differential Privacy, and hardware-based approaches enabled by trusted execution environments.”
“We had a substantial response to the call for papers for our workshop, with over 70 submissions from which the programme committee chose 35. Our workshop is sponsored by the Turing, alongside Amazon, Microsoft Research Cambridge, and Google, and the funding is helping us to support travel grants for international researchers which is a really important factor to making the workshop a success.”
Brooks Paige (University of Cambridge), Turing Research Fellow and co-organiser of Machine Learning for Molecules and Materials and Critiquing and Correcting Trends in Machine Learning, said:
"With my co-organisers Matt Kusner (Oxford, Turing) and José Miguel Hernández-Lobato (Cambridge, Turing), as well as colleagues at the Technical University of Berlin, we'll be running a workshop on machine learning for molecules and materials.”
“Molecular data and chemical processes provide serious challenges for machine learning methods, because even seemingly small changes to a molecule can have dramatic effects on its properties. This is really different from data like images, where changing a single pixel doesn't fundamentally alter the contents of an image. We’re trying to use machine learning in applications like predicting the outcomes of organic chemistry reactions, which is of central importance to the manufacture of medicines and materials, as well as for understanding many processes in molecular biology."
There are many other Turing researchers featured at NeurIPS 2018 – view the whole programme including workshops and the full list of accepted papers on the NeurIPS website.
 NeurIPS 2018 Opens; Best Papers Announced, Synced– 3 December 2018.